Flash Update

Current Presidential Model Results

Micro. Obama 318. Romney 220.
Macro. Obama 321. Romney 217.

Current Senate Model Results

Republican Seats
Median. 49
Most Frequent. 49

Probabilities

49+. 61%
50+. 41%
51+. 24%

NYT/CBS Q Polls Still Looking for Big Democratic Tilt

Looking Behind the Numbers in Florida and Ohio

Yesterday we examined a Virginia poll of likely voters from Virginia that was recently conducted jointly by the New York Times, Quinnipiac, and CBS News.  We noted that President Obama’s lead in that poll was predicated on an electorate that was even more heavily Democratic than indicated by 2008 exit polls.  In fact, if you recalculate top line results for that poll based off of the internal partisan indicators then Mitt Romney would be leading in Virginia unless the electorate is almost as Democratic as it was in 2008. 

Today, we’ll do an abbreviated version of that exercise for the new NYT/CBS/Q polls released for Florida, and Ohio.   WE ARE NOT ARGUING THAT THE PARTISAN MAKEUP OF LIKELY VOTERS IS DEFINITELY INCORRECT, BUT SUGGESTING THAT THEY RAISE INTERESTING QUESTIONS ABOUT HOW THE LIKELY VOTER MODEL IS CONSTRUCTED AND WHAT QUESTIONS THAT APPROACH RAISES.

Our view is that at the present time polling suggests that under a scenario where the electorate looks like 2008, President Obama would win if the election were held today.  If the electorate looks like 2004, then a Romney win is probable.

FLORIDA   NYT/CBS/Q Poll Obama 53 Romney 44

The following table pretty much says it all. In this poll Democrats hold a 9 percentage point edge in the pool of likely voters.  This compares to a Democratic advantage of 3 percentage points in 2008 AND a Republican advantage of 4 percentage points in 2004. 

Another way to look at this is the ratio of Democratic voters to Republican voters.   The poll suggests that there will be 1.33 Democratic voters for every Republican voter.  In 2008, the ratio was 1.09.  In 2004, it was 0.9.

Now this does not mean that President Obama is losing Florida.  It does suggest that a nine point advantage for Obama is highly unlikely here is what the race looks like at various partisan makeups.  The bottom line is that if Likely Voters were weighted to look more like 2004 than 2008, Romney would be leading in the poll.

A couple of other items jump out when looking at this poll’s internals.  While you have to be careful when looking at population subgroups based upon the sample size, here are a couple of interesting numbers that might seem inconsistent with the poll’s top line numbers.

·        Romney leads among Independents by a margin of 49%-46%.

·        Obama leads among Hispanics 55% - 41%.

·        Paul Ryan is viewed more favorably (38%-32%) by Independents than Joe Biden (37%-47%).

·        By a margin of 56% - 39%, Independent disapprove of President Obama’s job performance.

OHIO   NYT/CBS/Q Poll Obama 53 Romney 43

Similar to Florida, the composition of the electorate is critical to how you construct the pool of likely voters is critical.  In Ohio, a 2004 type electorate would move this in Romney’s direction.

This likely voter poll is slightly more Democratic than the 2008 electorate -- +9D vs. +8D.  The 2004 electorate was +5R.  The ratio of Democrats to Republicans in the poll was 1.35 to 1 vs. 1.26 in 2008 and 0.88 in 2004.  Small difference, but still more Democratic than 2008 voters.

Here are a couple of other interesting tidbits in the polls crosstabs.

 

·        Romney leads 47%-46% among Independents in this poll.

·        Obama leads by 60%-35% among women, Romney by 52-44% among men.

·        Among white voters Romney leads 49%-46%; 57%-40% with white men, and trails 53%-42% among white women.

Of course, party identification is not the only issue that can influence a poll’s outcome.  Other demographic factors such as gender, race, age and marital status can have a significant impact.  This is not to say the data behind the polls is wrong, but to remind readers that the interpretation of the data is also a major factor.

We won't know who is right until Election Day, but there is no question that the NYT/CBS/Q polls have a more generous view of this year's electorate for the President than even the 2008 turnout would justify.

Will the Obama Bubble Burst?

Are Prediction Markets Underestimating Extent of Obama’s Tail Risk?

Prediction markets like InTrade have proven to be good predictors of electoral outcome over the past several election cycles.  But the view six weeks prior to an election is not necessarily a good predictor.  In the last three weeks, there has been a frenzy of reporting on the momentum of the Obama campaign and the supposedly collapsing campaigns of Mitt Romney and several candidates for the United States Senate.

It is important to remember that prediction markets are just that – markets.  In the same way that polling provides a snapshot of where a particular race stands under a particular pollster’s assumptions, the market prices represent a snapshot of the prediction market’s collective wisdom at a particular point in time.  Just like markets for stocks and residential real estate, prediction markets are subject to what Alan Greenspan in 1996 referred to as “irrational exuberance.”  In that famous speech at the American Enterprise Institute preceding the bursting of the internet bubble and the housing bubble in the next decade, Greenspan asked “how do we know when irrational exuberance has unduly escalated asset values, which then become subject to unexpected and prolonged contractions …?”
Prediction markets generate important “wisdom of crowds” type data, but they are also subject to the “herd mentality” that can substitute objective analysis of underlying fundamentals with an “irrational exuberance.”  Just witness the reaction to rumors of leaked exit polls or unreported precincts and you will understand my point.

I am not saying that the current prices in the InTrade markets will be proven wrong.  I am suggesting that they may be driven by faulty assumptions about the fundamentals of the race.  Traders in these markets are avid followers of polls, rumors and blogs that purport to provide the latest wisdom and narrative.

Perhaps the most pervasive impact on prediction markets come from a poll driven narrative.  That means that polling failures, especially in endemic, will create significant market in the same manner that bad mortgage underwriting led to bad mortgages on over-valued homes or faulty corporate earnings reports led to inflated stock prices.

Let’s face facts.  It’s less likely that the polling data is wrong than it is that the pollsters’ models are wrong in the same way that bad underlying assumptions led S&P and Moody’s to rate packages of subprime mortgages as AAA.

The accuracy of polling in this cycle comes down to their accuracy in projecting the demographic makeup of the electorate.   And this is not just a question of looking at partisan divisions within a particular poll.  It comes down to an even broader set of demographic assumptions about age, sex, marital status, etc.

If you believe that the electorate in 2012 will be more demographically tilted toward President Obama than the 2008 election, then put your money on an Obama victory.  If you think the electorate will look like an average of the 2004 and 2008 electorate then you should be putting your money on Mitt Romney at this point in time. 
 
And as an aside, you should be extraordinarily skeptical of pollsters that just give the top lines from their poll and weave a narrative about the data, but don’t provide you with enough data to evaluate the degree to which their polling tilts one way or the other.

For instance, let’s examine the implications of how you answer the question of what the partisan makeup of the electorate.  The guinea pig for this exercise will be a recent NYT/CBS/Q poll of likely voters gives President Obama a 50%-46% lead in the battle for Virginia’s 13 electoral votes.  The poll numbers contribute to a media and pundit narrative that Mitt Romney is slipping badly or failing to gain traction in the Commonwealth.

A closer look at the polls internals suggest that Democrats could well end up having to call the Virginia electorate stupid.  The poll is based upon a likely voter composition that gives Democrats an 11 point advantage (35D 24R 36 I).  The pollsters could be right, but I’m skeptical.  In 2008, according to exit polls, Democrats enjoyed a partisan advantage of six points (39D 33R 27I).  So if you buy into the argument that the electorate in Virginia will be even more Democrat leaning than in 2008, trust the polls message, because that is what their likely voter sample is saying.

If you don’t buy that assumption, then consider the following table and chart.  The table illustrates what result would be generated using the poll’s internal partisan voting intentions and applying various electoral composition possibilities.   It looks at the 2008 and 2004 electoral compositions and the three points in between.  The result is that President Obama needs an electorate that is significantly more Democratic than the average of 2004 and 2008 to carry Virginia. 

A rough calculation suggests that in order for the President to carry Virginia he need a four points or better Democratic advantage in the electorate.  Why?  In this poll, Virginia independents favor Romney by 53%-42% -- an 11 point margin.  It’s pretty tough for anyone to win an election in Virginia if you lose independents by that margin.

The 2006 Senate exit polls gave Republicans a plus three advantage.  Democratic Senator Jim Webb won narrowly while carrying Independents by a 12 point margin.  I am hard pressed to recall any election in Virginia that was won by someone who lost Independents by double digits.  I’ll be really surprised if that happens in 2012.

State of the Races

Our micro model that examines State level data has President Obama winning a total of 312 electoral votes to Mitt Romney’s 236.  The macro level model gives President Obama a smaller margin of 302 – 236. 

The Republican prospects of taking control of the Senate have declined with the micro model presently projecting a median outcome of 49 seats for the Republicans.  The following chart shows the current probability distribution of Senate outcomes.